Diagnostic and Prognostic performance of blood plasma glycan … · 2019-02-26 · Recently, we...

1
Yueming Hu and Chad R. Borges* School of Molecular Sciences and The Biodesign Institute at Arizona State University, Tempe AZ 85287 Diagnostic and Prognostic performance of blood plasma glycan features in Women Epidemiology Lung Cancer (WELCA) study Our lab’s bottom-up “glycan node analysis” approach captures interesting glycan features such as “α2‐6 sialylation”, “β1‐4 branching” and “outer-arm fucosylation” as single analytical signals. From the WELCA study, plasma samples of 208 female lung cancer patients in stage I-IV and 207 age-matched healthy women were obtained. -----------------------------Introduction ----------------------------- Blood plasma and serum glycomics represents a promising source of new generation cancer biomarkers. Glycan node analysis, is a molecularly bottom-up approach to P/S glycomics developed by Borges et al. in 2013, focusing on monosaccharides and linkage specific “glycan nodes” rather than the intact glycan structures 1-2 . This approach captures all P/S glycans including N-, O-, and lipid-linked glycans and breaks them down into monosaccharides that maintains linkage information, by applying glycan linkage (methylation) analysis to whole biofluids ( Fig. 1). Furthermore, the important glycan features are captured and quantified as single analytical signals. In addition, many glycan nodes serve as direct surrogates for the activities of glycosyltransferases (GTs), enzymes that facilitate the construction of glycans and each of which is most commonly responsible for producing a unique glycan monosaccharide linkage pattern. Recently, we have applied glycan node analysis to several cancer studies, including pancreatic 3 , ovarian 3 , prostate 3 , breast 4 , and lung 1,3 cancer case-control study. For this study, 208 female patients with newly diagnosed stage I-IV lung cancer and 207 age-matched health women were obtained from Women Epidemiology Lung Cancer (WELCA) study. The purpose of this study was to further validate glycan node analysis as a means of detecting and predicting patient outcomes in lung cancer specifically in women. Interestingly, there are several important gender differences between men and women in lung cancer, including the facts that 1) after adjusting for the number of cigarettes smoked women have a three-fold greater risk of lung cancer than men; 2) never- smoker women are at significantly greater risk for lung cancer than men, and 3) women tend to have better survival rates than men. As such, we felt that for any differences observed in this study relative to our previously reported results in lung cancer 3 , it would also be important to look for any existing gender-based differences in glycan nodes as they may occur in the context of lung cancer. Figure 1 1 : Conceptual overview: An upregulated GT (e.g., GnT-V) causes an increase in the quantity of a specific, uniquely linked glycan monosaccharide residue (a 2,6-linked mannose “node” in this example)—which, through the subsequent action of other GTs, can lead to formation of a mixture of heterogeneous whole-glycan structures at low copy number each. Analytically pooling together the glycan “nodes” from amongst all the aberrant glycan structures in a given biomatrix provides a more direct surrogate measurement of GnT-V activity than any single intact glycan. Actual extracted ion chromatograms from 10-μL blood plasma samples shown. ----------------Conclusions/Summary------------------ ---------------------------Concept/Method--------------------------- --------------------------------------------------------Results -------------------------------------------------------- Diagnostic capacity of glycan features in lung cancer Dependence on smoking-status, age and histological type Ability to predict survival in lung cancer Early-stage detection performance Figure 2. Univariate distribution (a-d) and ROC curves (e-h) for the four top performing glycan nodes in the WELCA study. The Kruskal−Wallis test was performed followed by the Benjamini−Hochberg false discovery rate correction procedure. The letters at the top of data points in panels a-d demonstrate statistically significant differences between groups; any overlap in letters indicates a lack of significant differences between groups. ROC curves for stage I-IV lung cancer cases vs controls are provided in panels e-h. Areas under the ROC curves are provided in parenthesis next to the specified stages. “NS” next to the AUC values indicates that the ROC curve is not statistically significant. Figure 3. ROC curves for four top performing glycan nodes in early stages within different lung cancer sets. Four glycan nodes with highly-ranked performance in all three sample sets were shown. The ROC curves from WELCA sample set are illustrated in panels a,b. In panels c-e are ROC curves from the other two lung cancer sets: Stage I Only Lung Cancer set (c) and Dual Gender Lung Cancer set (d,e). Control: n = 207 Case: n = 16 Control: n = 73 Case: n = 107 Control: n = 199 Case: n = 20 Control: n = 207 Case: n = 13 Control: n = 199 Case: n = 20 Does Not Require Pre-Isolation of Proteins or Glycans Covers N-, O-, and Lipid-Linked Glycans Peak area normalized to internal standards (heavy glucose & heavy GlcNAc) O-glycans are released during permethylation. N-glycans and glycolipid glycans are released during acid hydrolysis. n values of each cohorts: Control: n = 207; Stage I: n = 16; Stage II: n = 13; Stage III: n = 45; Stage IV: n = 99. The results of Kruskal-Wallis test and ROC curve agree with our observations in a prior study 3 . a) e) b) c) f) g) d) h) a) b) c) d) e) a) c) b) d) a) b) c) Figure 4. The minor dependence on smoking status and age of the top performing glycan node 3,4-linked GlcNAc in the WELCA study. (a) The univariate distributions of outer-arm fucosylation within the control group are shown, subdivided by smoking status. Different letters above data points indicate statistically significant differences between groups by the Kruskal-Wallis test followed by the Benjamini-Hochberg FDR correction procedure. (b) Spearman’s rank correlation coefficients are provided above the data points. “**” next to the coefficient indicates p < 0.01. Controls are indicated by black triangles and cases by red dots. (c) The Kruskal−Wallis test was performed followed by the Benjamini−Hochberg FDR correction procedure to identify difference between age groups. For the other five top performing glycan nodes not shown in this figure, no statistically significant associations with age or smoking status were found. Glycan nodes/features were found to be independent of gender from two previous dual gender lung cancer sets. Therefore, a possible explanation for the better diagnostic performance of glycan features in early stages is the non-smoking matched control group involved in the WELCA set. Thus, dependence of glycan nodes/features on smoking-status, as well as age and histological type, was evaluated (Fig. 4). ROC curves of different stage IV histological types vs. controls were compared to each other by Delong’s test or Bootstrap test. No significant difference was found between histological types. The alteration of four glycan nodes (glycan features) in early stages of WELCA set were prominent compared to other two lung cancer sets 3 , which consisted of patients and controls in both genders. Figure 5. Survival curves for the four top performing glycan nodes for all stages combined. In each panel, the top quartile of specified glycan node is compared to all other quartiles combined. According to results of log- rank Mantel-Cox test, the survival curves within each panel are significantly different (p < 0.0001). Dotted lines represent 95% confident intervals. The median duration of follow-up for deceased patients (until death) was 406 days; for those that remained alive it was 1253 days. The median follow-up time for all patients was 1057 days. The top quartiles of all four glycan node markers predicted all-cause mortality with hazard ratios range from 2-3 and p < 0.01, by Cox proportional hazards regression model with adjustment for age, smoking status, and cancer stage. These results agree with our observations in a prior study 3 . Four glycan node-based features were able to separate lung cancer patients in nearly every stage from age-matched controls. Glycan features have minor or negligible dependence on smoking-status, age and histological type. The top quartiles of all four glycan node markers predicted all-cause mortality of lung cancer relative to all other quartiles combined. The above 3 conclusions validate findings observed in our previously published work 3 . Marked early-stage detection was observed in WELCA set compared to other two lung cancer sets. Gender did not appear to account for the improved early-stage diagnostic performance of glycan nodes in this study compared to our previously published work 3 . References [1] Borges CR, Rehder DS, Boffetta P. 2013. Analytical chemistry 85: 27-36 [2] Zaare S, Aguilar J, Hu Y, Ferdosi S, Borges C. 2016. Journal of visual29ized experiments: JoVE [3] Ferdosi S, Rehder DS, Maranian P, Castle EP, Ho TH, et al. 2018. Journal of Proteome Research 17: 543-58 [4] Hu Y, Borges CR. 2017. Analyst 142: 2748-59

Transcript of Diagnostic and Prognostic performance of blood plasma glycan … · 2019-02-26 · Recently, we...

Page 1: Diagnostic and Prognostic performance of blood plasma glycan … · 2019-02-26 · Recently, we have applied glycan node analysis to several cancer studies, including pancreatic 3,

Yueming Hu and Chad R. Borges*School of Molecular Sciences and The Biodesign Institute at Arizona State University, Tempe AZ 85287

Diagnostic and Prognostic performance of blood plasma glycan features in Women Epidemiology Lung Cancer (WELCA) study

▪ Our lab’s bottom-up “glycan node analysis” approach captures interesting glycan

features such as “α2‐6 sialylation”, “β1‐4 branching” and “outer-arm fucosylation” as

single analytical signals.

▪ From the WELCA study, plasma samples of 208 female lung cancer patients in stage I-IV

and 207 age-matched healthy women were obtained.

-----------------------------Introduction -----------------------------

Blood plasma and serum glycomics represents a promising source of new generation cancer

biomarkers. Glycan node analysis, is a molecularly bottom-up approach to P/S glycomics developed

by Borges et al. in 2013, focusing on monosaccharides and linkage specific “glycan nodes” rather

than the intact glycan structures 1-2. This approach captures all P/S glycans including N-, O-, and

lipid-linked glycans and breaks them down into monosaccharides that maintains linkage

information, by applying glycan linkage (methylation) analysis to whole biofluids (Fig. 1).

Furthermore, the important glycan features are captured and quantified as single analytical signals.

In addition, many glycan nodes serve as direct surrogates for the activities of glycosyltransferases

(GTs), enzymes that facilitate the construction of glycans and each of which is most commonly

responsible for producing a unique glycan monosaccharide linkage pattern.

Recently, we have applied glycan node analysis to several cancer studies, including pancreatic 3,

ovarian 3, prostate 3, breast 4, and lung 1,3 cancer case-control study. For this study, 208 female

patients with newly diagnosed stage I-IV lung cancer and 207 age-matched health women were

obtained from Women Epidemiology Lung Cancer (WELCA) study. The purpose of this study was to

further validate glycan node analysis as a means of detecting and predicting patient outcomes in

lung cancer specifically in women. Interestingly, there are several important gender differences

between men and women in lung cancer, including the facts that 1) after adjusting for the number

of cigarettes smoked women have a three-fold greater risk of lung cancer than men; 2) never-

smoker women are at significantly greater risk for lung cancer than men, and 3) women tend to

have better survival rates than men. As such, we felt that for any differences observed in this study

relative to our previously reported results in lung cancer 3, it would also be important to look for any

existing gender-based differences in glycan nodes as they may occur in the context of lung cancer.

Figure 1 1: Conceptual overview: An upregulated GT (e.g., GnT-V)causes an increase in the quantity of a specific, uniquely linkedglycan monosaccharide residue (a 2,6-linked mannose “node” inthis example)—which, through the subsequent action of other GTs,can lead to formation of a mixture of heterogeneous whole-glycanstructures at low copy number each. Analytically pooling togetherthe glycan “nodes” from amongst all the aberrant glycan structuresin a given biomatrix provides a more direct surrogatemeasurement of GnT-V activity than any single intact glycan. Actualextracted ion chromatograms from 10-µL blood plasma samplesshown.

----------------Conclusions/Summary------------------

---------------------------Concept/Method---------------------------

--------------------------------------------------------Results --------------------------------------------------------

Diagnostic capacity of glycan features in lung cancer Dependence on smoking-status, age and histological type

Ability to predict survival in lung cancer

Early-stage detection performance

Figure 2. Univariate distribution (a-d) and ROC curves (e-h) for the four top performing glycan nodes in the WELCA study. The Kruskal−Wallis test was performed followed by the Benjamini−Hochberg false discovery rate correction procedure. The letters at the top of data points in panels a-d demonstrate statistically significant differences between groups; any overlap in letters indicates a lack of significant differences between groups. ROC curves for stage I-IV lung cancer cases vs controls are provided in panels e-h. Areas under the ROC curves are provided in parenthesis next to the specified stages. “NS” next to the AUC values indicates that the ROC curve is not statistically significant.

Figure 3. ROC curves for four top performing glycan nodes in early stages within different lung cancer sets. Four glycan nodes with highly-ranked performance in all three sample sets were shown. The ROC curves from WELCA sample set are illustrated in panels a,b. In panels c-e are ROC curves from the other two lung cancer sets: Stage I Only Lung Cancer set (c) and Dual Gender Lung Cancer set (d,e).

Control: n = 207Case: n = 16

Control: n = 73Case: n = 107

Control: n = 199Case: n = 20

Control: n = 207Case: n = 13

Control: n = 199Case: n = 20

▪ Does Not Require Pre-Isolation of Proteins or Glycans

▪ Covers N-, O-, and Lipid-Linked Glycans

▪ Peak area normalized to internal standards (heavy glucose & heavy GlcNAc)

O-glycans are released during permethylation.

N-glycans and glycolipid glycans are released during acid hydrolysis.

▪ n values of each cohorts:

▪ Control: n = 207; Stage I: n = 16; Stage II: n = 13; Stage III: n = 45; Stage IV: n = 99.

▪ The results of Kruskal-Wallis test and ROC curve agree with our observations in a prior study 3.

a) e)

b)

c)

f)

g)

d) h)

a)

b)

c)

d)

e)

a)

c)

b)

d)

a) b) c)

Figure 4. The minor dependence on smoking status and age of the top performing glycan node 3,4-linked GlcNAc in the WELCA study. (a) The univariate distributions of outer-arm fucosylation within the control group are shown, subdivided by smoking status. Different letters above data points indicate statistically significant differences between groups by the Kruskal-Wallis test followed by the Benjamini-Hochberg FDR correction procedure. (b) Spearman’s rank correlation coefficients are provided above the data points. “**” next to the coefficient indicates p < 0.01. Controls are indicated by black triangles and cases by red dots. (c) The Kruskal−Wallis test was performed followed by the Benjamini−Hochberg FDR correction procedure to identify difference between age groups. For the other five top performing glycan nodes not shown in this figure, no statistically significant associations with age or smoking status were found.

▪ Glycan nodes/features were found to be independent of gender from two previous dual gender lung cancer sets. Therefore, a possible explanation for the better diagnostic performance of glycan features in early stages is the non-smoking matched control group involved in the WELCA set. Thus, dependence of glycan nodes/features on smoking-status, as well as age and histological type, was evaluated (Fig. 4).

▪ ROC curves of different stage IV histological types vs. controls were compared to each other by Delong’s test or Bootstrap test. No significant difference was found between histological types.

▪ The alteration of four glycan nodes (glycan features) in early stages of WELCA set were prominent compared to other two lung cancer sets 3, which consisted of patients and controls in both genders. Figure 5. Survival curves for the four top performing glycan nodes for all stages combined. In each panel, the

top quartile of specified glycan node is compared to all other quartiles combined. According to results of log-

rank Mantel-Cox test, the survival curves within each panel are significantly different (p < 0.0001). Dotted lines

represent 95% confident intervals. The median duration of follow-up for deceased patients (until death) was 406

days; for those that remained alive it was 1253 days. The median follow-up time for all patients was 1057 days.

▪ The top quartiles of all four glycan node markers predicted all-cause mortality with hazard ratios range from 2-3 and p < 0.01, by Cox proportional hazards regression model with adjustment for age, smoking status, and cancer stage. These results agree with our observations in a prior study 3.

➢ Four glycan node-based features were able to separate lung cancer patients in nearly

every stage from age-matched controls.

➢ Glycan features have minor or negligible dependence on smoking-status, age and

histological type.

➢ The top quartiles of all four glycan node markers predicted all-cause mortality of lung

cancer relative to all other quartiles combined.

➢ The above 3 conclusions validate findings observed in our previously published work 3.

➢ Marked early-stage detection was observed in WELCA set compared to other two lung

cancer sets.

➢ Gender did not appear to account for the improved early-stage diagnostic performance

of glycan nodes in this study compared to our previously published work 3.

References

[1] Borges CR, Rehder DS, Boffetta P. 2013. Analytical chemistry 85: 27-36[2] Zaare S, Aguilar J, Hu Y, Ferdosi S, Borges C. 2016. Journal of visual29ized experiments: JoVE[3] Ferdosi S, Rehder DS, Maranian P, Castle EP, Ho TH, et al. 2018. Journal of Proteome Research 17: 543-58[4] Hu Y, Borges CR. 2017. Analyst 142: 2748-59